Dataset for 'Existing hydropower can firm large low-carbon capacity for datacenters'
『既存水力発電はデータセンター向け大規模低炭素電源として活用可能』のためのデータセット (AI 翻訳)
Cheng Feng, Fengqi You
🤖 gxceed AI 要約
日本語
本研究のデータセットは、既存水力発電所がデータセンターに対して確実な低炭素電源を供給できる可能性を示した論文を支えるものです。全球81地域のSWAT+モデル、LSTM-Transformerによる機械学習流量予測、4193ユニットの水力発電所データベースが含まれ、AIを用いた水力発電の柔軟性評価を可能にします。
English
This dataset supports the study showing that existing hydropower can provide firm low-carbon capacity for datacenters. It includes SWAT+ models for 81 global regions, machine-learning streamflow predictions using LSTM-Transformer, and a database of 4,193 hydropower units, enabling AI-assisted assessment of hydropower flexibility.
Unofficial AI-generated summary based on the public title and abstract. Not an official translation.
📝 gxceed 編集解説 — Why this matters
日本のGX文脈において
日本のデータセンター電力需要増加と再生可能エネルギー拡大において、既存水力発電の活用は有力な選択肢です。本データセットは日本を含む全球地域をカバーしており、国内の水力発電所の柔軟性評価に応用可能です。
In the global GX context
Globally, datacenters' growing electricity demand requires firm low-carbon sources. This dataset provides a replicable framework for assessing how existing hydropower can fulfill that role, relevant to ISSB and TCFD reporting on climate resilience and renewable energy integration.
👥 読者別の含意
🔬研究者:Provides a comprehensive global dataset for modeling hydropower firming capacity with AI, useful for energy system modelers.
🏢実務担当者:Datacenter operators can use these methods to evaluate hydropower as a reliable low-carbon energy source, reducing carbon footprint.
🏛政策担当者:Highlights the potential of existing hydropower for grid reliability and low-carbon datacenter growth, informing energy policy.
📄 Abstract（原文）
================================================================================Full reproduction-data archive"Existing hydropower can firm large low-carbon capacity for datacenters"================================================================================ This is the LARGE data tier of the study. It is the companion to the public coderepository, which holds all analysis/model code plus a small (~135 MB) bundle thatalready reproduces every figure on its own: Code + figure-reproduction bundle: https://github.com/VictorCFeng/Hydropower This archive holds the heavy and one-of-a-kind data that does not fit in the coderepository. Reference climate combination throughout: GFDL-ESM4 / SSP3-7.0,existing hydropower only ("nofuture"), with the full 15-member GCM x SSP ensemblewhere a result is inherently multi-model. --------------------------------------------------------------------------------CONTENTS-------------------------------------------------------------------------------- swat_global/ [~630 GB; NOT on GitHub] The complete per-region modeling tree for all 81 grid regions, comprising: * the built SWAT+ models (QSWAT+ project, sqlite database, HRU / channel / routing definitions, watershed shapefiles and DEM rasters) -- everything needed to re-run SWAT+ for a region; * the trained machine-learning streamflow models: 4,023 per-channel LSTM-Transformer probabilistic models (LSTM_Transformer_prob/); * the machine-learning-corrected hourly streamflow that drives the dispatch, for all 15 GCM x SSP combinations (Scenarios/isimip3b/<GCM>/<SSP>/ and inference_output/.../corrected_all.parquet); * the per-region datasets, vectors, rasters and figures used to build them. WEATHER FORCING IS EXCLUDED on purpose (.pcp/.tmp/.hmd/.slr/.wnd, ~16 GB per region). It is fully regenerable with the ISIMIP3b / ERA5-Land download scripts in stages 01-02 of the code repository. station_screening/ [~51 MB; NOT on GitHub] 90 per-region "rivers and plants" screening maps -- the visual record of how each hydropower station was matched onto the modeled river network during the station-screening step. hydropower_station_database.csv [~0.75 MB; NOT on GitHub] The 4,193 dispatched hydropower units inside the modeled grid regions: unified identity, capacity, location, river, dam height, the net head used for dispatch (final_head_m) with its confidence and source, the head-storage-area rating coefficients, and the GRFR routing reach each unit is matched to. figures/ [~13 MB] The full set of 64 rendered figure panels. (The code repository ships the 25 paper panels; this is the complete set, including supplementary and alternate-metric variants.) --------------------------------------------------------------------------------NOT INCLUDED (used upstream; obtain from the original providers)-------------------------------------------------------------------------------- * Weather forcing (ISIMIP3b / ERA5-Land) -- regenerable via the stage-01/02 download scripts in the code repository. * S&P Global "451 Research Datacenter KnowledgeBase" -- commercial, licensed; only aggregated results derived from it appear in the study. * IUCN Red List of Threatened Species -- licensed; the conservation overlay (paper Fig. 4e/4f) is regenerable from an IUCN download with the included fig5_*.py scripts. --------------------------------------------------------------------------------CONTACT / LICENCE-------------------------------------------------------------------------------- Licence: data CC BY 4.0 (code is MIT, in the repository above) Contact: [email protected] (PEESE Group, Cornell University)
🔗 Provenance — このレコードを発見したソース
- openalex https://doi.org/10.5281/zenodo.21285281first seen 2026-07-13 05:43:30
🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。
gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。